Extreme Rainfall in NYC from
Ida: Insights from a High-
Resolution Forecast Model
Ty Janoski1,2, James Booth1, and Thomas Galarneau2
1 NOAA-CESSRST-II, City College of New York, New York, NY
2NOAA OAR National Severe Storms Laboratory, Norman, OK
Background & Motivation
Urban Areas are Susceptible to Flash
Flooding
Permeable
soil
Vegetation
to intercept
and absorb
water
Only ~10%
runoff
Impervious
concrete &
asphalt
Little
vegetation
55% runoff
on average
Stormwater Drainage System
The Nature Conservancy
Hourly Rainfall in NYC
Hourly rainfall main metric
for flash flood risk
NYC infrastructure designed for a
5-year storm (20% chance any
year)
Threshold: 1.75 inches per hour
Statistica
Central Park ASOS Hourly Rainfall Amounts
Note: These values are from only one weather
station and regular hourly METARs
Consider these lowball measurements.
2021 was already anomalously wet
2021: NYCs second wettest
summer on record
21 August 2021: Hurricane
Henri brings 1.94”/hr
Central Park ASOS Hourly Rainfall Amounts
Statistica
New York Post
Ida shatters previous records
Central Park ASOS Hourly Rainfall Amounts
NYC Micronet Rainfall Measurements
Statistica
NYS Mesonet
Impact: Flash Flooding
Hours-long flash flooding
13 dead in NYC
11 from basement flooding
Billions of dollars of
damage
Different from past storms:
affected inland
neighborhoods instead of
coasts
Adapted from Fig. 1 of Mossel et al. (2024)
Events like Ida have and will become
more common
Mossel et al. (2024) found that
the return period of Ida-like
events has decreased over past
4 decades
Trend likely to continue and
accelerate with climate change
Despite increasing likelihood,
Ida-like events are super rare
Return Period (inverse of likelihood) of Ida-like hourly rainfall
Adapted from Fig. 2 of Mossel et al. (2024)
Meteorological History
What exactly happened?
Hurricane Ida Track & Transition
Formed: Aug 26, 2021
ET Transition: Sep 1, 2021
Dissipated: Sep 4, 2021
NOAA National Hurricane Center Source: Hart & Evans
Hurricane Ida Dated Track & Classifications
Hurricane Ida Phase Space Diagram
Upper-level support
Ohio Valley trough @ 250 mb
Right-entrance region of jet
streak = UL Divergence =
Ascent = Precipitation!
UL divergence also supports a
low-level jet to bring energy
and moisture
Right-entrance region
of jet streak
Mid-level Support
Surface Analysis
Warm front
responsible for
heavy NYC rainfall
Mixed-Layer CAPE & Precipitable Water
Mixed-
Layer
CAPE is a
measure
of
instability
Precipitable
water is a
measure of
moisture
Radar Loop and 24 hr Rainfall
MRMS Q3 Multisensor 24-hour Rainfall
KOKX Radar & NWS Watches/Warnings
Valid 2030 UTC 1 Sep 20210310 UTC 2 Sep 2021
Period Ending 0900 UTC 2 Sep 2021
Summary & what’s next
The confluence of many factors across spatial scales made Ida so
destructive
Trough and jet streak placement
Positive vorticity advection
Low-level jet
Slow-moving front
Ample moisture and instability
Training supercells
Which of these offers the most predictability for Ida and future
extreme rainfall events?
Results with the Warn-on-
Forecast System
Model: Warn-on-Forecast System (WoFS)
18-member ensemble analysis and forecast
system
3 km horizontal resolution; 51 vertical levels
Based on WRF-ARW
Movable domain!
Data assimilation every 15 minutes
Radar (WSR-88D)
Satellite (GOES)
Surface observations
6-hour-long simulations initialized hourly 1700
0300 UTC
Physics and initial conditions (EnKF) differ
between members
Radar locations ( ) with 150 km
range rings
WoFS Grid on 1 Sep 2021
2100 Initialization of WoFS
Paintball plot
of reflectivity
values > 40
dBZ
Green = flash
flood warning
Red = Tornado
warning
Blue = Severe
Thunderstorm
Warning
Region and rainfall metric
WoFS Output: Hourly
initializations from 20002300
UTC 1 Sep 2021
4 initializations × 18 members =
72-member ensemble
Rainfall metric: 01000200*UTC
rainfall in NYC = RAIN1-2,NYC
Corresponds to observed
maximum rainfall rates in
Manhattan
Metric region over
NYC and
surrounding areas
NYC” = region where 72-member
ensemble SD of hourly rainfall ≥65%
of domain-maximum SD
*Results are
relatively insensitive
to hour chosen
Huge ensemble spread in NYC rainfall!
Few ensemble
members come
close to
matching
observations!
Approach #1: Compare
the 5 wettest and 5
driest ensemble
members from each
initialization → Wet &
Dry Subsets (n=20)
Histograms of the 0100-0200 UTC NYC Rainfall in each WoFS initialization
Dry subset has less rain, shifted NW
Probability Matched Mean*of 01000200 UTC Rainfall
*Probability
matched
mean →
similar to
ensemble
mean, but
retains
extreme
values
Dotted = statically
significant
Wet subset has higher maximum rainfall
over entire WoFS domain
Histogram of the Max 0100-0200 UTC Rainfall over
all ensemble members and WoFS domain
Ensemble Sensitivity Analysis
Ensemble Sensitivity Analysis = a linear regression of a forecast metric
against the atmosphere at an earlier time

 󰇛󰇜
󰇛󰇜 󰇛󰇜
Here,  (0100-0200 UTC NYC Rainfall) is our forecast
metric, while is our predictor (geopotential heights, winds, etc.)
Resulting values = expected change in  given a one-
standard deviation change in
Sensitivity Slope of linear regression StDev of
predictor
Sensitivity of RAIN1-2,NYC to Geopotential Heights
500 mb Heights
850 mb Heights
2300 UTC 0000 UTC 0100 UTC
Likely a sign
of existing
convection,
not synoptic
scale
circulation
differences!
Higher heights
to N and SW
impact steering
flow direction
and speed of
supercells!
I only show values that are
statistically significantly
different from 0!
850 mb Zonal Wind
850 mb Meridional Wind
2300 UTC 0000 UTC 0100 UTC
Sensitivity of RAIN1-2,NYC to 850 mb Winds
Cyclonic shear↑ =
frontogenesis↑ =
Ascent↑ = Rain
Stronger low-level
jet = more
moisture and
instability
Precipitable Water
Mixed-Layer CAPE
2300 UTC 0000 UTC 0100 UTC
Precipitable water
to the south↑ =
RAIN1-2,NYC
Mixed-Layer
CAPE↑ = More fuel
for convection
CAPE: Convective Available
Potential Energy (Instability)
Sensitivity of RAIN1-2,NYC to PW and CAPE
Let’s look at smaller (meso,
frontal) scale features
Back to wet vs. dry!
Hodographs
Hodographs show how
wind changes with height
Wet subset has slower,
more eastward moving
supercells
Our first clue that the front
location is different in wet
vs. dry!
Dashed = Dry Subset Mean
Solid = Wet Subset Mean
Polar coordinates:
θ = Wind direction
r = Wind Speed
Blue arrows are an
estimate of storm
motion
2300 UTC 0000 UTC 0100 UTC
Wet & Dry UH > 20m2s-2 with Trajectory Angles
Wet
Dry
Wet subset has
greater number
and trajectory
angle of
supercells
Dry subset
missing Jersey
Shore storms
UH = Updraft helicity, a
measure of “spininess” to
track supercells!
Front Finding Algorithm
Created a new front-finding algorithm for WoFS using the thermal
front parameter (TFP) (Schemm et al. 2015)
  

where  is the equivalent potential temperature at 850 hPa
Fronts occur where TFP = 0 and is above a certain threshold
Use connected component labeling to find the largest feature, which
is the front!
2300 UTC 0000 UTC 0100 UTC
WoFS Front Locations
Warm front in wet subset located
farther south than in dry subset!
Take-home Messages
Hurricane Ida flooding in NYC resulted from a confluence of
factors to produce record-breaking hourly rainfall
It was a tough forecast, with huge spread among WoFS members
The WoFS members that produced the most rain in NYC have:
A synoptic setup that favors slower, more eastward moving supercells
A stronger low-level jet to bring moisture and instability
A stronger warm front located farther south in NJ
These factors should be considered
References
Mossel, C., Hill, S. A., Samal, N. R., Booth, J. F., & Devineni, N. (2024).
Increasing extreme hourly precipitation risk for New York City after
Hurricane Ida. Scientific Reports, 14(1), 27947.
https://doi.org/10.1038/s41598-024-78704-9
Schemm, S., Rudeva, I., & Simmonds, I. (2015). Extratropical fronts in
the lower troposphereglobal perspectives obtained from two
automated methods. Quarterly Journal of the Royal Meteorological
Society, 141(690), 16861698. https://doi.org/10.1002/qj.2471